Space: evaluation history 1.00-1.04 (PNG+CSV) - Latarnik 14e70755, GREEN Wartownik 02:28Z

.gitattributes

@@ -34,3 +34,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
 scores-history.png filter=lfs diff=lfs merge=lfs -text
+scores-history-104.png filter=lfs diff=lfs merge=lfs -text

index.html

@@ -150,8 +150,8 @@
 
 <section>
   <h2>Historical score curves: 1.00–1.02</h2>
-  <figure><img src="https://huggingface.co/spaces/SlayerLab/Slayer1B/resolve/b67cc3f1c3a56ea16f2ae92f44023e70750e371d/scores-history.png" alt="Evaluation history of Slayer1B 1.00, 1.01 and 1.02 versus training tokens"></figure>
-  <p class="note">Measured values only, no fits: one checkpoint every 150 steps. Historical data: <a href="https://huggingface.co/spaces/SlayerLab/Slayer1B/resolve/b67cc3f1c3a56ea16f2ae92f44023e70750e371d/scores-history.csv">scores-history.csv</a>.</p>
+  <figure><img src="scores-history-104.png" alt="Evaluation history of Slayer1B 1.00, 1.01, 1.02, 1.03 and 1.04 versus training tokens"></figure>
+  <p class="note">Measured values only, no fits. 1.00-1.02: one checkpoint every 150 steps. 1.04: the in-run evaluator at 50-step cadence, same metric definitions, preliminary (the run continues, no learning-rate decay yet); its series has a gap 8850-10010 caused by a pipeline restart. 1.03 was measured with the older lm-eval task set, so it appears only in the loss panel (dotted line marks its end); its composite "eff" is not computed here. Data: <a href="scores-history.csv">scores-history.csv</a> (1.00-1.02) and <a href="scores-history-104.csv">scores-history-104.csv</a> (1.04).</p>
 </section>
 
 <section>
@@ -326,8 +326,8 @@
 
 <section>
   <h2>Historyczne krzywe wyników: 1.00–1.02</h2>
-  <figure><img src="https://huggingface.co/spaces/SlayerLab/Slayer1B/resolve/b67cc3f1c3a56ea16f2ae92f44023e70750e371d/scores-history.png" alt="Historia wyników Slayer1B 1.00, 1.01 i 1.02 w funkcji liczby tokenów treningu"></figure>
-  <p class="note">Same pomiary, bez dopasowań: punkt kontrolny co 150 kroków. Historyczne dane: <a href="https://huggingface.co/spaces/SlayerLab/Slayer1B/resolve/b67cc3f1c3a56ea16f2ae92f44023e70750e371d/scores-history.csv">scores-history.csv</a>.</p>
+  <figure><img src="scores-history-104.png" alt="Historia wyników Slayer1B 1.00, 1.01, 1.02, 1.03 i 1.04 w funkcji liczby tokenów treningu"></figure>
+  <p class="note">Same pomiary, bez dopasowań. 1.00-1.02: punkt kontrolny co 150 kroków. 1.04: ewaluator w biegu, kadencja 50 kroków, te same definicje metryk, wstępne (bieg trwa, bez wygaszania LR); seria ma przerwę 8850-10010 po restarcie pipeline'u. 1.03 mierzono starszym zestawem lm-eval, więc jest tylko na panelu straty (kropkowana linia oznacza jego koniec); złożonego „eff" tu nie liczymy. Dane: <a href="scores-history.csv">scores-history.csv</a> (1.00-1.02) i <a href="scores-history-104.csv">scores-history-104.csv</a> (1.04).</p>
 </section>
 
 <section>

scores-history-104.csv

@@ -0,0 +1,38 @@
+model,step,tokens,arc_easy,blimp,wikitext2_byte_ppl,multiblimp_pl,arc_easy_pl,eval_gpu,note
+1.04,7200,7549747200,49.75,74.63,2.3439437403551735,98.41,43.1,JP-A100,"preliminary, run in progress, 50-step in-run evaluator"
+1.04,7250,7602176000,49.87,74.7,2.3455666117624507,98.47,42.55,JP-A100,"preliminary, run in progress, 50-step in-run evaluator"
+1.04,7300,7654604800,49.2,74.09,2.3351618122876463,98.56,43.14,JP-A100,"preliminary, run in progress, 50-step in-run evaluator"
+1.04,7350,7707033600,49.75,74.71,2.3388517725163487,98.59,43.22,JP-A100,"preliminary, run in progress, 50-step in-run evaluator"
+1.04,7400,7759462400,48.65,74.9,2.338034606486071,98.66,42.59,JP-A100,"preliminary, run in progress, 50-step in-run evaluator"
+1.04,7450,7811891200,49.96,74.44,2.341254921553752,98.53,43.27,JP-A100,"preliminary, run in progress, 50-step in-run evaluator"
+1.04,7500,7864320000,48.99,74.11,2.338087919610084,98.17,42.93,JP-A100,"preliminary, run in progress, 50-step in-run evaluator"
+1.04,7550,7916748800,49.16,75.2,2.3375954100344454,98.59,43.64,JP-A100,"preliminary, run in progress, 50-step in-run evaluator"
+1.04,7600,7969177600,48.95,75.1,2.3377119179564523,98.66,43.39,JP-A100,"preliminary, run in progress, 50-step in-run evaluator"
+1.04,7650,8021606400,48.99,75.21,2.3393352270951033,98.78,43.14,JP-A100,"preliminary, run in progress, 50-step in-run evaluator"
+1.04,7700,8074035200,49.87,74.94,2.3381095402003615,98.72,43.22,JP-A100,"preliminary, run in progress, 50-step in-run evaluator"
+1.04,7750,8126464000,48.65,74.23,2.334415446531806,98.59,42.97,JP-A100,"preliminary, run in progress, 50-step in-run evaluator"
+1.04,7800,8178892800,49.2,74.61,2.342433981523328,98.62,44.11,JP-A100,"preliminary, run in progress, 50-step in-run evaluator"
+1.04,7850,8231321600,48.36,74.7,2.330855481463633,98.81,43.01,JP-A100,"preliminary, run in progress, 50-step in-run evaluator"
+1.04,7900,8283750400,48.53,75.28,2.336526498194587,98.72,42.93,JP-A100,"preliminary, run in progress, 50-step in-run evaluator"
+1.04,7950,8336179200,50.08,75.06,2.333592621419013,98.84,43.48,JP-A100,"preliminary, run in progress, 50-step in-run evaluator"
+1.04,8000,8388608000,49.71,74.77,2.336944683312258,98.9,43.06,JP-A100,"preliminary, run in progress, 50-step in-run evaluator"
+1.04,8050,8441036800,48.86,74.14,2.3349375810439486,98.78,42.85,JP-A100,"preliminary, run in progress, 50-step in-run evaluator"
+1.04,8100,8493465600,48.95,74.06,2.3334274001978965,98.87,42.26,JP-A100,"preliminary, run in progress, 50-step in-run evaluator"
+1.04,8150,8545894400,50.04,74.58,2.329613679374312,98.78,43.14,JP-A100,"preliminary, run in progress, 50-step in-run evaluator"
+1.04,8200,8598323200,49.37,73.96,2.323286197648698,98.9,43.27,JP-A100,"preliminary, run in progress, 50-step in-run evaluator"
+1.04,8250,8650752000,49.71,74.09,2.333923090468432,98.93,43.64,JP-A100,"preliminary, run in progress, 50-step in-run evaluator"
+1.04,8300,8703180800,50.13,74.32,2.3333254477388885,98.59,43.77,JP-A100,"preliminary, run in progress, 50-step in-run evaluator"
+1.04,8350,8755609600,49.71,74.83,2.3314559809920903,98.75,43.94,JP-A100,"preliminary, run in progress, 50-step in-run evaluator"
+1.04,8400,8808038400,50.93,75.3,2.3232721847845172,98.5,42.97,JP-A100,"preliminary, run in progress, 50-step in-run evaluator"
+1.04,8450,8860467200,49.79,74.39,2.323663562203663,98.66,43.52,JP-A100,"preliminary, run in progress, 50-step in-run evaluator"
+1.04,8500,8912896000,50.34,74.49,2.3292563368471595,98.53,43.94,JP-A100,"preliminary, run in progress, 50-step in-run evaluator"
+1.04,8550,8965324800,49.54,74.52,2.331142499523616,98.56,43.73,JP-A100,"preliminary, run in progress, 50-step in-run evaluator"
+1.04,8600,9017753600,49.79,74.56,2.3262990629840363,98.66,43.77,JP-A100,"preliminary, run in progress, 50-step in-run evaluator"
+1.04,8650,9070182400,49.62,75.18,2.322438099701948,98.56,44.44,JP-A100,"preliminary, run in progress, 50-step in-run evaluator"
+1.04,8700,9122611200,49.54,74.84,2.322443450988014,98.66,42.93,JP-A100,"preliminary, run in progress, 50-step in-run evaluator"
+1.04,8750,9175040000,50.25,74.85,2.3301453905353333,98.66,42.51,JP-A100,"preliminary, run in progress, 50-step in-run evaluator"
+1.04,8800,9227468800,51.52,75.38,2.3211318404351533,98.72,43.69,JP-A100,"preliminary, run in progress, 50-step in-run evaluator"
+1.04,8850,9279897600,50.13,75.19,2.320471997848369,98.35,43.64,JP-A100,"preliminary, run in progress, 50-step in-run evaluator"
+1.04,8900,9332326400,49.92,75.36,2.3227954918686526,98.47,43.86,JP-A100,"preliminary, run in progress, 50-step in-run evaluator"
+1.04,8950,9384755200,49.71,75.36,2.323378618531867,98.56,43.64,JP-A100,"preliminary, run in progress, 50-step in-run evaluator"
+1.04,10010,10496245760,50.59,74.17,2.3077690263752513,98.87,44.02,JP-A100,"preliminary, run in progress, 50-step in-run evaluator"